Lightnews — Scholar-powered news

Blaine @blaine.bsky.social · 14h

Otoh, I once built a functional XMPP PubSub implementation from a 50 line profile of XEP60, so maybe details aren't all that? 💃

Blaine @blaine.bsky.social · 14h

oh, right, gotcha. Yeah, this is the "broken" part of my brain that just glosses over those details and says "what you _mean_ is..." This is [part of] why I've never been fun at standards parties. 😂

1

Blaine @blaine.bsky.social · 15h

I also love that there's a lot of convergence (aka rough consensus in running code), AND excited about the spots where there's divergence, because it means there's creativity and simultaneous invention, which are early signs of a promising direction.

2

Blaine @blaine.bsky.social · 15h

... rather than *just* having a generic s2s delivery endpoint with user targeting. 🤷‍♀️ Mysteries of the universe! 😅

1

Blaine @blaine.bsky.social · 15h

Yeahhhh; I mean, user@server == server@server[user1,...,userN] if it's direct s2s delivery. Again, I don't know the specs without looking (and don't have time to research, sadly) but I am surprised to learn that ActivityPub implements user inboxes as individual endpoints, ...

1

Blaine @blaine.bsky.social · 15h

I love that things that we're building internally at <Super Not Secret New_ Public Project> are being validated daily in the wider Atmosphere.

This is [roughly] how we're handling pre-moderation, and I'm impressed daily by how easy it is to add behaviours to our system with atproto.

Nick Gerakines @ngerakines.me · 2d

Built cryptographically-verified endorsements for at://work. The core innovation isn’t just cryptography, it’s the consent model. Both parties must explicitly agree to create a verified attestation. This changes how trust networks form. @atwork.place #ATProtocol

Building Unforgeable Professional Endorsements with ATProtocol - Nick's Blog

Traditional professional endorsements on platforms like LinkedIn lack cryptographic proof—anyone could forge them, and the platform controls the truth. This article introduces a two-record architectur...

ngerakines.leaflet.pub

1 2 16

Blaine @blaine.bsky.social · 15h

If only someone had invented a mechanism to list multiple recipients on a message envelope!

Full disclosure: I haven't looked at the specs on this, so this is just an uninformed shitpost.

1

Blaine @blaine.bsky.social · 2d

Yeah, implanting LLMs matrix-style would be Bad Times. 😂

It's way better for figuring out how to pump water from a stream to an outlet 600m away than YouTube, though.

To be fair, implanting YouTube matrix-style... that's not even gonna reach the scifi cutting room floor. *twitch*

2

Blaine @blaine.bsky.social · 2d

(fwiw, I've never been able to get into podcasts, either - it's not just you 😅)

2 2

Blaine @blaine.bsky.social · 2d

Also, the paid models are much better. One of the better takes I've heard is that we need to re-evaluate our assumptions about LLMs every 3 months, given the rate of change (for better and worse!). Also, fuck billionaires, obviously, which probably sours the whole LLM pot. 😢

1 2

Blaine @blaine.bsky.social · 2d

For research, same thing; I usually want to understand a thing that's either widely understood (e.g., specific applications of fluid dynamics) or that I understand but want to dig into nuance without spending ages searching and skimming (with constant risk of ADD interruption).

1 2

Blaine @blaine.bsky.social · 2d

Yeah, makes sense! Fwiw, in code I mostly *want* "boring mainstream views"; at least for what I'm doing now, clever software engineering is an anti-goal. Where the weird bits are, that's me and I'm not sure an LLM would take anyone else to the same conclusions.

1 1

Blaine @blaine.bsky.social · 3d

Is it perfect? No. Is it better than me spending hours piecing together a similar-but-worse understanding and probably not writing it down very well, if at all? Hell yes.

1 3

Blaine @blaine.bsky.social · 3d

I also use (mostly) ChatGPT to do broad analysis of things that would take hours of research. E.g.,

"Give me a detailed description of how the OAuth system works in Bluesky/atproto, along with historical context of what pitfalls in the old system led to the new one being developed."

1 3

Blaine @blaine.bsky.social · 3d

It also turns my ADD into a superpower. I can happily operate 4-5 claude sessions working on totally different things in parallel, in ways that I think would break a lot of neurotypical people.

1 5

Blaine @blaine.bsky.social · 3d

I can also leave prompts with context for days, and come back to them without needing to spend hours spinning back up on my own context, which is *huge* for ADD-brain, because that's often the hardest part.

2 3

Blaine @blaine.bsky.social · 3d

I've been using Claude basically 100% of the time. Hard to summarize, but I act at the level of "architect" or "principal engineer" and guide the system towards solutions, but don't have to worry about being an expert in the finer details of eg how jest interacts with expo's build system.

1 4

Blaine @blaine.bsky.social · 3d

I'm steadfast in my belief that LLM coding is an assistive aid. It's like having a 24x7 pairing partner for ADD-raddled minds like mine. I wish I had the focus of the people who say it's shit, but ... I don't.

2 1 14

Blaine @blaine.bsky.social · 3d

So much this. I worked on a research project on this topic with @inkandswitch.com a few years back: www.inkandswitch.com/upwelling/

Even though I love where we got to, we barely scratched the surface; there's so much room for improvement.

Upwelling: Combining real-time collaboration with version control for writers.

Collaborative writing tools don’t work well for writers or editors. With Upwelling, we demonstrate a design that gives writers privacy while still offering editors transparency into how a document is ...

www.inkandswitch.com

1 3

Blaine @blaine.bsky.social · 3d

I feel this in my bones.

Mark Hurrell @markhurrell.bsky.social · 3d

the class divide in tech between research, strategy and making things has always been this way www.theguardian.com/world/2025/o...

3

Blaine @blaine.bsky.social · 3d

That was pretty fun! 😄

1

Blaine @blaine.bsky.social · 4d

wrote a reply on the post, but for folks looking here, my vote is for the community DID approach. Feels natural and is (was!) easy to implement.

bnewbold.leaflet.pub/3m2x7bilyrc2...

AT Namespaces for Community Spaces - little emanations

How should public community-controlled spaces be represented in atproto?

bnewbold.leaflet.pub

7

Blaine @blaine.bsky.social · 10d

A.I. at its peak

A screenshot of the titles scene from Serial Experiments: Lain of Lain, with eyes closed, wearing a hat with a cute bear face on it.

6

Blaine @blaine.bsky.social · 11d

The science appears to suggest that this is precisely what's happening. youtu.be/CYlon2tvywA?...

If only we had lightly connected small world networks instead of highly connected big world ones... 😅

Something Strange Happens When You Trace How Connected We Are

YouTube video by Veritasium

youtu.be

2 1 8

Blaine @blaine.bsky.social · 11d

At that point of a project where I'm giving things clever names.

1 15